Current jailbreak attacks on large language models (LLMs) predominantly rely on LLMs themselves to generate adversarial prompts, creating a critical efficiency bottleneck: each attack requires substantial computational resources and API queries, limiting scalability and practical deployment. To overcome this limitation, we propose Adversarial Prompt Distillation (APD), a novel framework that transfers jailbreaking capabilities from LLMs to small language models (SLMs) for efficient, low-resource attacks. APD integrates three key components: (1) masked adversarial knowledge pre-training via LoRA fine-tuning, (2) dynamic temperature-controlled knowledge distillation to bridge architectural gaps, and (3) reinforcement learning-based template optimization for adaptive refinement. Extensive experiments across 12 models show that APD achieves state-of-the-art attack success rates (e.g., 96.4% ASR_k on GPT-4) while dramatically improving efficiency—generating prompts 3.7× faster with 11.3× fewer parameters than teacher models. Our work establishes the first practical framework for lightweight jailbreak attacks, exposes new vulnerabilities in LLM defenses, and provides a scalable testbed for advancing AI safety research. Our code is available at: https://github.com/lxgem/Efficient_and_Stealthy_Jailbreak_Attacks_via_Adversarial_Prompt.

Language agents are increasingly used as social simulators, but most are evaluated by local utterance plausibility rather than their ability to reproduce longitudinal discourse dynamics. We introduce PRISM, a calibrated language-agent framework for forecasting and counterfactual simulation of social media threads. PRISM separates three components that are often conflated in prompt-only agents: latent affective evolution, participation control, and language generation. It models affect with a profile-conditioned stochastic differential equation, selects participation through a partially observed activation policy, and uses a language model as a controlled generator conditioned on the resulting state. Agent profiles are instantiated with MBTI-like anchors, but treated only as behavioral priors rather than psychological ground truth. On 47,207 post-rooted discussion threads spanning 159 trending topics and 356,593 hierarchical comments, PRISM outperforms homogeneous, random, continuous-trait, prompt-only, and memory-only baselines in trajectory alignment, held-out thread forecasting, and blind human evaluation. Reliability analysis further validates LLM-derived weak labels against human and independent-model annotations. These results suggest that calibrated state-space agents offer a more reliable foundation for online discourse simulation than persona prompting alone.

Large Language Models (LLMs) are increasingly equipped with capabilities of real-time web search and integrated with protocols like the Model Context Protocol (MCP). This extension could introduce new security vulnerabilities. We present a systematic investigation of LLM vulnerabilities to hidden adversarial prompts through malicious font injection in external resources like webpages, where attackers manipulate code-to-glyph mapping to inject deceptive content which are invisible to users. We evaluate two critical attack scenarios: (1) malicious content relay and (2) sensitive data leakage through MCP-enabled tools. Our experiments reveal that indirect prompts with injected malicious font can bypass LLM safety mechanisms through external resources, achieving varying success rates based on data sensitivity and prompt design. Our research underscores the urgent need for enhanced security measures in LLM deployments when processing external content.

Recent multimodal large language models (MLLMs) have demonstrated significant potential in open-ended conversation, generating more accurate and personalized responses. However, their abilities to memorize, recall, and reason in sustained interactions within real-world scenarios remain underexplored. This paper introduces MMRC, a Multi-Modal Real-world Conversation benchmark for evaluating six core open-ended abilities of MLLMs: information extraction, multi-turn reasoning, information update, image management, memory recall, and answer refusal. With data collected from real-world scenarios, MMRC comprises 5,120 conversations and 28,720 corresponding manually labeled questions, posing a significant challenge to existing MLLMs. Evaluations on 20 MLLMs in MMRC indicate an accuracy drop during open-ended interactions. We identify four common failure patterns: long-term memory degradation, inadequacies in updating factual knowledge, accumulated assumption of error propagation, and reluctance to “say no.” To mitigate these issues, we propose a simple yet effective NOTE-TAKING strategy, which can record key information from the conversation and remind the model during its responses, enhancing conversational capabilities. Experiments across six MLLMs demonstrate significant performance improvements.